Non-IID and aware federated intrusion detection with PBFT with secured model aggregation for multi institutional healthcare internet of things networks

Description

Sengan and Shieh propose, in 2026, a federated intrusion detection system for hospital Internet of Things networks spanning many sites. Patient data cannot legally be pooled there, yet attacks must be caught jointly. Their design confronts two problems at once. Each hospital's data distribution differs, the non-IID problem that cripples federated averaging. And some participants may be Byzantine, submitting poisoned updates to corrupt the model.

Hospitals are clustered by how alike their data distributions are, using a Wasserstein-distance affinity matrix, with gradient divergence correction before a Byzantine-tolerant consensus protocol secures the combining step. On test traffic the system reached accuracies of 93.17 percent on the IoT-Flock dataset and 89.84 percent on CICIoT2023 in non-IID conditions. It still held 83.61 percent accuracy with four Byzantine attackers, converged within 29 to 67 rounds, and met a 16-millisecond real-time constraint the authors set. They frame it as ready protection for device-dense wards.

Catalog record

Source: https://europepmc.org/articles/PMC13415842